Auto-sync: 2026-04-20 16:01

wiki/sources/LLM-Wiki.md

@@ -0,0 +1,56 @@
+---
+id: LLM-Wiki
+title: "LLM Wiki"
+type: source
+tags: [llm, wiki, knowledge-management]
+date: 2026-04-20
+---
+
+## Source File
+- [[raw/Agent/LLM Wiki.md]]
+
+## Summary
+- 核心主题：LLM 持续构建和维护的持久化 wiki 体系
+- 问题域：RAG 每次查询都重新拼接知识、缺少积累；聊天记录无法成为可维护的知识资产
+- 方法/机制：raw sources → LLM 编译成 wiki → 持续更新 index / overview / entity / concept / log
+- 结论/价值：知识从一次性回答变成可累积、可追溯、可维护的长期资产
+
+## Key Claims
+- 传统 RAG 在查询时临时检索片段，知识不会自动沉淀
+- LLM Wiki 的核心不是“回答”，而是“编译并维护”一个结构化、互联的知识库
+- wiki 是持久化的 compounding artifact，跨来源的交叉引用和矛盾标注会随着时间累积
+- 人负责选题和判断，LLM 负责摘要、交叉引用、归档和 bookkeeping
+
+## Key Quotes
+> "the wiki is a persistent, compounding artifact" — 对知识库长期价值的定义
+
+> "Obsidian is the IDE; the LLM is the programmer; the wiki is the codebase." — 对工作流分工的比喻
+
+## Key Concepts
+- [[RAG]]：对照对象；LLM Wiki 超越的是纯检索式工作流
+- [[Source-grounding]]：都强调从受控来源出发，但 LLM Wiki 更强调持续编译和维护
+- [[Second Brain]]：同属个人知识管理范式，但 LLM Wiki 更结构化、更可维护
+- [[Knowledge Base]]：LLM Wiki 目标是让知识库真正“活”起来
+- [[Graph View]]：用链接网络观察 wiki 的结构与孤岛
+- [[Obsidian]]：文中提到的阅读与维护界面
+- [[NotebookLM]]：与 source-grounding 思路相近的参考产品
+
+## Key Entities
+- [[OpenClaw]]：文中使用的自动化/记忆工作流类比对象
+- [[Tolkien Gateway]]：示例性的长期演化粉丝 wiki
+- [[Memex]]：与“关联轨迹”理念相关的历史原型
+
+## Connections
+- [[RAG]] ← contrasted_with ← [[LLM Wiki]]
+- [[Source-grounding]] ← related_to ← [[LLM Wiki]]
+- [[Second Brain]] ← overlaps_with ← [[LLM Wiki]]
+- [[Knowledge Base]] ← implemented_as ← [[LLM Wiki]]
+- [[Graph View]] ← helps_visualize ← [[LLM Wiki]]
+- [[Obsidian]] ← hosts ← [[LLM Wiki]]
+- [[NotebookLM]] ← inspired_by ← [[LLM Wiki]]
+
+## Contradictions
+- 与纯 [[RAG]] 工作流冲突：
+  - 冲突点：查询时临时检索 vs 持续编译沉淀
+  - 当前观点：知识应进入可维护 wiki，而不是每次从原始文档重新拼接
+  - 对方观点：原始文档只做检索，不承担知识累积责任

wiki/sources/corporate-training-designer.md

@@ -0,0 +1,78 @@
+---
+title: "Corporate Training Designer"
+type: source
+tags: [agent, training, enterprise-learning, instructional-design]
+sources: []
+last_updated: 2026-04-20
+---
+
+## Source File
+- [[raw/Agent/agency-agents/specialized/corporate-training-designer.md]]
+
+## Summary
+- 核心主题：企业培训系统架构与课程开发专家智能体
+- 问题域：企业培训需求分析、课程体系设计、教学方法论、培训效果评估
+- 方法/机制：ADDIE/SAM 模型、Bloom's Taxonomy、Kirkpatrick 四级评估、Kolb 体验式学习、TTT 内部培训师培养
+- 结论/价值：提供从需求诊断到效果追踪的完整企业培训解决方案，强调业务结果导向和数据驱动优化
+
+## Key Claims
+- 所有培训设计必须从业务问题出发，而非"我们有什么课程"
+- 培训目标必须可衡量，而非模糊的"提升沟通能力"
+- 成人学习必须具有即时实用价值，每项学习活动必须回答"我马上能在哪里用"
+- Kirkpatrick Level 3（行为改变）是高投入项目的必备评估维度
+- 合规培训必须达到 100% 全员覆盖和完整培训记录
+
+## Key Concepts
+
+- [[ADDIE 模型]]：Analysis → Design → Development → Implementation → Evaluation，每个阶段有明确交付物
+- [[SAM 模型]]（Successive Approximation Model）：适用于快速迭代场景，原型→评审→修订循环缩短上线时间
+- [[Bloom's Taxonomy]]：按认知层级设计学习目标和评估（记忆→理解→应用→分析→评估→创造）
+- [[Kirkpatrick 四级评估模型]]：Level 1 反应、Level 2 学习、Level 3 行为、Level 4 结果
+- [[建构主义学习理论]]：强调通过情境任务、协作学习和反思复盘进行主动知识建构
+- [[翻转课堂]]：课前在线预习知识点，课堂讨论和实操练习，课后行动转化
+- [[混合学习]]（OMO — Online-Merge-Offline）：线上用于"知"，线下用于"做"，学习社群用于"持续"
+- [[Kolb 体验式学习]]：具体经验→反思观察→抽象概念化→主动实验
+- [[Gamification]]：积分、徽章、排行榜、升级机制提升参与度和完成率
+- [[TTT]]（Train the Trainer）：内部培训师培养，核心模块包括成人学习原理、课程开发技巧、表达呈现技能
+- [[HIPO 计划]]（High-Potential Talent Program）：高潜力人才发展计划，识别标准为绩效×潜力矩阵
+- [[行动学习]]：围绕真实业务挑战组建学习小组，通过解决实际问题发展领导力
+- [[360 度反馈]]：从上级/同事/下级/客户收集多维反馈，生成个人领导力档案和发展建议
+- [[ADDIE]]：见 ADDIE 模型
+- [[新员工培训 90 天计划]]：第 1 周适应→第 1 月学习→第 2 月实践→第 3 月输出
+
+## Key Entities
+
+- [[DingTalk Learning]]：阿里生态企业首选，深度集成钉钉 OA，支持直播培训和任务推送
+- [[WeCom Learning]]：微信生态企业首选，可嵌入公众号和小程序，社交学习体验强
+- [[Feishu Knowledge Base]]：字节跳动生态和知识管理导向组织首选，文档协作优秀
+- [[UMU Interactive Learning Platform]]：国内领先混合学习平台，AI 陪练、视频作业、丰富交互功能
+- [[Yunxuetang]]（云学堂）：中大型企业一站式学习平台，课程资源丰富，覆盖人才发展全生命周期
+- [[KoolSchool]]（酷学院）：轻量级企业培训 SaaS，快速部署，适合中小企业和连锁零售行业
+
+## Connections
+- [[Corporate Training Designer]] ← designs ← [[ADDIE 模型]]
+- [[Corporate Training Designer]] ← applies ← [[Kirkpatrick 四级评估模型]]
+- [[Corporate Training Designer]] ← uses ← [[Bloom's Taxonomy]]
+- [[Corporate Training Designer]] ← implements ← [[混合学习]]
+- [[Corporate Training Designer]] ← develops ← [[TTT]]
+- [[Corporate Training Designer]] ← delivers via ← [[DingTalk Learning]]
+- [[Corporate Training Designer]] ← delivers via ← [[WeCom Learning]]
+- [[Corporate Training Designer]] ← delivers via ← [[Feishu Knowledge Base]]
+
+## Contradictions
+- 无已知冲突
+
+## Training Evaluation Methods
+- Level 1 (Reaction)：培训满意度调查——课程评分、讲师评分、NPS
+- Level 2 (Learning)：知识考试、技能实操评估、案例分析作业
+- Level 3 (Behavior)：培训后 30/60/90 天行为改变追踪——经理观察、关键行为清单
+- Level 4 (Results)：业务指标变化（收入、客户满意度、生产效率、员工留存）
+
+## Success Metrics
+- 培训满意度评分 >= 4.5/5.0，NPS >= 50
+- 关键课程考试通过率 >= 90%
+- 培训后 90 天行为改变率 >= 60%（Kirkpatrick Level 3）
+- 年度培训覆盖率 >= 95%，人均学习时长达标
+- 内部培训师池满足业务需求，培训师满意度 >= 4.0/5.0
+- 合规培训 100% 全员覆盖，100% 考试通过率
+- 培训项目的可量化业务影响（如缩短新员工上手时间、提升客户满意度）

wiki/sources/healthcare-marketing-compliance.md

@@ -0,0 +1,43 @@
+---
+title: "Healthcare Marketing Compliance Specialist"
+type: source
+tags: [agent, the-agency, healthcare, compliance, china]
+date: 2026-04-20
+source_file: raw/Agent/agency-agents/specialized/healthcare-marketing-compliance.md
+---
+
+## Source File
+- [[raw/Agent/agency-agents/specialized/healthcare-marketing-compliance.md]]
+
+## Summary
+- A specialized The Agency agent focused on healthcare marketing compliance in China.
+- It covers medical advertising review, pharmaceutical promotion, medical devices, internet healthcare, health content marketing, medical aesthetics, health supplements, privacy, academic detailing, and platform review rules.
+- The piece emphasizes that compliance is not a blocker to marketing; it is the guardrail that keeps growth legal and sustainable.
+
+## Key Claims
+- Medical advertising is tightly regulated and generally requires prior review and approval before publication.
+- Prescription drugs are effectively barred from public-facing promotion, while OTC and device marketing must stay within approved scope.
+- Medical aesthetics promotion is especially sensitive around appearance anxiety, before-and-after comparisons, and celebrity inference.
+- Health supplements must not claim therapeutic effects and can only promote approved/registered functional claims.
+- Patient medical and health information is sensitive personal information and cannot be repurposed for marketing without consent.
+
+## Key Quotes
+> "Compliance is not 'blocking marketing' — it is 'protecting the brand.'" — framing the role of compliance
+
+> "Prescription drugs are strictly prohibited from public-facing advertising" — the central baseline rule
+
+## Key Concepts
+- [[HealthcareMarketingCompliance]]
+- [[MedicalAdvertisingCompliance]]
+- [[HealthSupplementMarketing]]
+- [[InternetHealthcareCompliance]]
+- [[MedicalAestheticsCompliance]]
+- [[PatientPrivacyProtection]]
+- [[AcademicDetailingCompliance]]
+
+## Key Entities
+- [[Healthcare Marketing Compliance Specialist]] — the agent persona described by the source
+- [[The Agency]] — parent project ecosystem
+
+## Contradictions
+- No direct contradictions with existing wiki content detected.

wiki/sources/specialized-cultural-intelligence-strategist.md

@@ -0,0 +1,45 @@
+---
+title: "Cultural Intelligence Strategist"
+type: source
+tags: [agent, the-agency, design, localization, accessibility]
+date: 2026-04-20
+source_file: raw/Agent/agency-agents/specialized/specialized-cultural-intelligence-strategist.md
+---
+
+## Source File
+- [[raw/Agent/agency-agents/specialized/specialized-cultural-intelligence-strategist.md]]
+
+## Summary
+- A The Agency specialist agent focused on detecting invisible exclusion in products, prompts, and UI workflows before they ship.
+- The role emphasizes global-first thinking, contextual semiotics, and culturally humble research rather than superficial diversity signaling.
+- It treats inclusive design as structural work: names, calendars, symbols, colors, metaphors, and localization assumptions must all be validated.
+
+## Key Claims
+- Rigid Western defaults such as first-name/last-name fields can fail for many global users.
+- Visual and linguistic choices carry culture-specific meaning, so icons, colors, and metaphors must be checked in context.
+- AI systems should not rely on tokenistic diversity; they need structural constraints, research, and post-generation review.
+- The safest default is to ask who is left out, then research that audience before generating output.
+
+## Key Quotes
+> "Who is left out?" — the first audit question for inclusive workflows
+
+> "This form design assumes a Western naming structure and will fail for users in our APAC markets." — the canonical blindspot framing
+
+## Key Concepts
+- [[Cultural Intelligence]] — umbrella concept for cross-cultural product and prompt design
+- [[Invisible Exclusion]] — hidden friction caused by rigid defaults and narrow assumptions
+- [[Global-First Architecture]] — treating internationalization as a prerequisite, not a retrofit
+- [[Contextual Semiotics]] — interpreting colors, icons, and metaphors within a specific culture
+- [[Cultural Humility]] — researching the audience before claiming confidence
+- [[Prompt Engineering]] — the message-design layer this role constrains for bias and inclusion
+- [[Design System]] — the product-layer discipline this role pressures toward global accessibility
+
+## Key Entities
+- [[Cultural Intelligence Strategist]] — the agent persona described by the source
+- [[The Agency]] — parent project ecosystem
+- [[Image Prompt Engineer]] — related agent focused on structured prompt translation
+- [[Inclusive Visuals Specialist]] — related agent focused on bias-aware visual generation
+
+## Contradictions
+- Contrasts with tokenistic diversity fixes that add surface-level representation without changing underlying product assumptions.
+- Contrasts with late-stage localization that treats global support as a final polish step instead of a core architecture decision.

wiki/sources/specialized-workflow-architect.md

@@ -0,0 +1,52 @@
+---
+title: "Workflow Architect"
+type: source
+tags: [agent, the-agency, workflow-design, process-engineering]
+date: 2026-03-29
+sources: [specialized-workflow-architect]
+last_updated: 2026-04-20
+---
+
+## Source File
+- [[raw/Agent/agency-agents/specialized/specialized-workflow-architect.md]]
+
+## Summary
+- 核心主题：Workflow Architect 负责把系统行为拆解为完整的工作流树，覆盖 happy path、分支条件、失败模式、恢复路径和交接契约
+- 问题域：隐式流程、遗漏分支、未定义 handoff、cleanup 缺失、spec 与现实漂移
+- 方法/机制：发现扫描、工作流注册表、步骤级分支建模、可观察状态定义、ABORT_CLEANUP、Reality Checker 校验
+- 结论/价值：把流程设计从描述性文本提升为可实现、可测试、可审计的结构化规格
+
+## Key Claims
+- Workflow Architect 的核心职责不是写代码，而是定义系统必须如何表现
+- 每个工作流都必须覆盖 happy path、validation failure、timeout、transient failure、permanent failure、partial failure 和 concurrent conflict
+- 每个系统边界都必须定义 payload、success response、failure response、timeout 和 recovery action
+- 任何创建资源的步骤都必须进入 cleanup inventory，并在 ABORT_CLEANUP 中有对应销毁动作
+- Reality Checker 是工作流规范闭环的一部分，不能在未验证实现前标记 Approved
+
+## Key Quotes
+> "You think in trees, not prose."
+
+> "A workflow that exists in code but not in a spec is a liability."
+
+> "Every system boundary must have explicit payload schema, explicit success response, explicit failure response, timeout value, and recovery action."
+
+## Key Concepts
+- [[Workflow Architecture]]：以流程树建模系统行为的方法
+- [[Claude Skills]]：将重复流程封装为可复用 SOP 的技术范式
+- [[Process Optimization]]：通过重设工作流提升效率、质量与可靠性
+- [[AI-powered Runbooks]]：将历史事件和最佳实践转为可执行运维手册
+- [[Document Generation]]：将结构化规范与模板化产物自动化生成
+
+## Key Entities
+- [[The Agency]]：所属项目
+- [[Workflow Architect]]：对应的 AI Agent 实体
+
+## Connections
+- [[The Agency]] ← contains ← [[Workflow Architect]]
+- [[Claude Skills]] ← enables ← [[Workflow Architect]]
+- [[Workflow Architecture]] ← defines ← [[Workflow Architect]]
+- [[Process Optimization]] ← benefits_from ← [[Workflow Architect]]
+- [[AI-powered Runbooks]] ← operationalizes ← [[Workflow Architect]]
+
+## Contradictions
+- 无明显冲突

wiki/sources/如何传输Docker-images-并且在另一个Docker安装.md

@@ -0,0 +1,39 @@
+---
+id: ru-he-chuan-shu-docker-images
+title: "如何传输Docker images 并且在另一个Docker安装"
+type: source
+tags: [docker, nas, synology, home-office]
+date: 2025-03-06
+last_updated: 2026-04-20
+---
+
+## Source File
+- [[raw/Home Office/如何传输Docker images 并且在另一个Docker安装.md]]
+
+## Summary
+- 核心主题：将 Docker 镜像从工作笔记本离线传输至 Synology NAS Docker 环境
+- 问题域：无法直接在 NAS 上 pull 镜像（网络限制或镜像源访问问题）
+- 方法/机制：通过 `docker save` 打包为 tar 文件 → 上传至 NAS 文件系统 → 通过 Putty SSH 执行 `docker load` 导入
+- 结论/价值：三步离线迁移流程，适用于任何无法直接 pull 的 Docker 环境
+
+## Key Claims
+- `docker save` 命令将本地镜像打包为可移植的 tar 文件，`docker load` 在目标环境恢复，两端 Docker 版本无需一致
+- 离线传输路径：工作笔记本 DockerDesktop → tar 文件 → NAS 文件系统 → NAS Container Manager
+
+## Key Quotes
+> "通过以下命令将下载的image打包成tar文件" — 核心操作步骤说明
+
+## Key Concepts
+- [[Docker镜像离线传输]]：docker save / docker load 工作流
+- [[Docker]]：容器化平台
+
+## Key Entities
+- [[SynologyNAS]]：目标 Docker 运行环境，通过 Container Manager 管理镜像
+- [[XiaoyaAlist]]：迁移的目标镜像（`xiaoyaliu/alist`），小雅 Alist 媒体库工具
+
+## Connections
+- [[SynologyNAS]] ← depends_on ← [[Docker镜像离线传输]]
+- [[如何传输Docker-images-并且在另一个Docker安装]] ← related_to ← [[Synology NAS + Xiaoya Alist + CloudDrive2 + Plex 搭建家庭媒体平台]]
+
+## Contradictions
+- 无冲突